The problem
What was broken before AI
Teams that needed lots of on-brand visuals usually had to brief designers from scratch, reuse a limited asset library, or prompt a general-purpose model and manually fight style drift. Even when a prompt described the right vibe, the model could miss details that matter to a brand system: stroke weight, color palette, lighting, character features, or the feel of a specific photographic set.
What changed
What the use case made possible
Firefly Custom Models gives the team a way to upload a curated set of owned images, review AI-generated metadata, confirm rights and permissions, train a custom model, and then generate new images from that model inside Firefly or Firefly Boards. Adobe’s own help documentation says the beta supports illustration style, photographic style, and character use cases, with 10–30 JPG or PNG images required for training.
Why this matters
Why this use case is worth studying
This case is useful because it shifts the creative AI conversation from prompt cleverness to asset governance. A brand does not only need a better sentence; it needs a permissioned training set, a repeatable review process, and a model that encodes the visual rules people already use. That makes the AI workflow closer to maintaining a design system than asking for one-off images.
Use this when
When this pattern applies
Use this when a team needs many visual variations that should still feel like they came from the same brand, artist, campaign, character universe, or photographic system.
